lone-parent families, provide few sucessful role models. This set of variables is treated by means
of standard factorial ecology methods. We first ran a Principal Component Analysis to define
a number of non-correlated factors summarizing the information carried by these variables (see
Table A.1 in Appendix). Then, we gathered neighborhoods according to their respective coordi-
nates on the factorial axes using a hierarchical ascending classification method (with the Ward
method that minimizes intra-group variance). We obtained five13 clusters of neighborhoods that
are presented in Appendix (Table A.2).
In order to deal with a dummy variable, we grouped the least two favored neighborhood
types as opposed to the rest of the city, thus defining the endogenous variable y2 . These neigh-
borhoods, labelled “deprived” in the rest of the paper, represent 35% of the 540 neighborhoods
and of the population of Lyon’s city. They are spread in different parts of the city, still mostly
concentrated in its eastern half (Figure 3). They are characterized by high unemployment rates
(twice as high as the average unemployment rate of other neighborhoods), high percentages of
foreigners and low educational levels and professional statuses (Table 2). Most of them have a
large share of public housing, but 10% of them have less than 10% of public housing units.
3.3 Neighborhood, public housing and unemployment: descriptive statistics
Table 3 provides a few sample statistics by neighborhood type and by whether the individual is
renter in the public sector or not. Deprived neighborhoods host almost one third of the individ-
uals in our sample. Among deprived neighborhoods, 41% of individuals are renters in the public
sector, against only 9% in other neighborhoods. Other residents in deprived neighborhoods are
either renters in the private sector or homeowners (33% and 61% of them respectively). About
one third of public housing renters in our sample are located in neighborhoods that are not
classified as deprived. Thus, the diversity of situations regarding the combination of tenures and
neighborhood types allows us to disentangle the effect of the two residential variables.
Compared with individuals having the same tenure (public housing versus others), in-
dividuals in deprived neighborhoods are less educated and have lower occupational statuses.
Yet, they have similar demographic characteristics, except for public housing renters in deprived
neighborhoods who have larger families than their counterparts in the rest of the city, owing to
a large share of foreign families having more children than the average.
13This was the optimal number of clusters, according to a wide variety of criteria, including the Cubic Clustering
Criterion, Pseudo-F and Pseudo-t values.
12